Cornelis Nets $205 Million to Build AI Networking Gear in Challenge to Nvidia’s Interconnect Grip

Networking startup Cornelis said Monday it has raised $205 million in new funding as it works on technology designed to let AI chips exchange data more efficiently — a piece of the computing stack where NVIDIA has long held a commanding position.

As artificial intelligence systems scale to thousands of processors working in tandem, the links between chips have become as important as the chips themselves. NVIDIA pairs its accelerators with its own interconnect and networking products, reinforcing its lead in the data center. Cornelis is among a wave of companies building alternative networking hardware aimed at reducing that lock-in.

Shares of NVIDIA, which operates as a data center-scale AI infrastructure company through its Compute & Networking and Graphics segments, traded at $223.67 on Monday, down 0.73% from the prior close of $225.31. The chipmaker’s market capitalization stands at roughly $5.56 trillion.

Cornelis did not detail in its announcement how it plans to deploy the fresh capital, but companies in this space typically direct funding toward engineering headcount, silicon development, and early customer deployments. The raise signals continued investor appetite for the infrastructure layer of the AI buildout, where enormous training clusters have created demand for faster, more efficient data movement between accelerators, storage, and servers.

The competitive landscape is crowded. Beyond NVIDIA’s in-house networking portfolio, hyperscalers and emerging AI hardware vendors have backed Ethernet-based and open-standard approaches to interconnects, hoping to lower costs and avoid dependence on a single supplier. Startups like Cornelis argue that purpose-built networking can squeeze more performance out of large clusters regardless of which accelerator sits at the node.

The $205 million round ranks among the more sizable recent funding events in AI infrastructure networking, a category that has drawn increasing attention as model training runs have grown to consume tens of thousands of processors.

What to watch

• Disclosure of the funding round’s lead investors and valuation, if Cornelis provides details
• Announcements of design wins or deployments with AI cloud providers and data center operators
• NVIDIA’s upcoming earnings report, where networking and interconnect revenue trends may offer a read on competitive dynamics
• Broader adoption of open networking standards across large-scale AI training clusters

Source: original release